Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Invariant Graph Propagation in Constraint-Based Local Search

Authors: Frej Knutar Lewander, Pierre Flener, Justin Pearson

JAIR 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental the measuring of the throughput (number of probes per second) of the algorithms for various invariant graphs, validating our recommendations (Section 4). We developed a CBLS solver, called Atlantis, that can propagate an invariant graph in both input-to-output and output-to-input styles. For the output-to-input style, the solver supports the total, ad-hoc, and prepared marking strategies. We ran our experiments on a desktop computer with an ASUS PRIME Z590-P motherboard, a 3.5 GHz Intel Core i9 11900K processor, and four 16 GB 3200 MT/s DDR4 memories, running Ubuntu 22.04.4 LTS with GCC (the GNU Compiler Collection) 11. The results are shown in Figure 9.
Researcher Affiliation Academia FREJ KNUTAR LEWANDER , PIERRE FLENER, and JUSTIN PEARSON, Uppsala University, Sweden. Authors Contact Information: ... Uppsala University, Department of Information Technology, Uppsala, Sweden.
Pseudocode Yes Algorithm 1: The propagation of an invariant graph in output-to-input style. Algorithm 2: The propagation of an invariant graph in input-to-output style.
Open Source Code Yes The source code of Atlantis is publicly available at https://github.com/astra-uu-se/atlantis/
Open Datasets No The difficulty and realism of the problem instances are thus unimportant, so we generated random parameter values instead of retrieving instances from existing repositories.
Dataset Splits No For each invariant graph model, we generated 9 instances, of sizes 16, 32, 64, 96, 128, 196, 256, 512, and 1024. The paper does not mention any training/test/validation splits, as it focuses on performance measurement rather than machine learning model evaluation.
Hardware Specification Yes We ran our experiments on a desktop computer with an ASUS PRIME Z590-P motherboard, a 3.5 GHz Intel Core i9 11900K processor, and four 16 GB 3200 MT/s DDR4 memories, running Ubuntu 22.04.4 LTS with GCC (the GNU Compiler Collection) 11.
Software Dependencies Yes running Ubuntu 22.04.4 LTS with GCC (the GNU Compiler Collection) 11.
Experiment Setup Yes For each invariant graph model, we generated 9 instances, of sizes 16, 32, 64, 96, 128, 196, 256, 512, and 1024. The throughput is measured for each instance when the input-to-output propagation style (denoted input-to-output ), output-to-input propagation style with ad-hoc marking (denoted output-to-input ad-hoc ), output-to-input propagation style with prepared marking (denoted output-to-input prepared ), and output-to-input propagation style with total marking (denoted output-to-input total ) are used.